Neural Network Feature Map Segmentation for Mobile Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Machine learning models for object detection in images are typically large and resource-intensive, making them unsuitable for devices with limited processing power and memory, such as smartphones, and struggle to provide real-time results in applications like augmented reality.

Innovation Solution

A compact machine-learning model architecture optimized for efficiency, using a neural network with a region proposal network and convolutional layers to process images in a way that allows for real-time object detection and segmentation on devices with limited resources, by generating feature maps and combining regional feature maps into larger tensors for efficient convolutional operations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a large machine learning model is used for object detection, then detection accuracy is improved, but device resource consumption increases

Engineering Contradiction:
Improvedetection accuracyVSAvoiddevice resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent divides the object detection task into multiple stages: region proposal generation, region of interest selection, and final detection. This segmentation allows the system to process only relevant regions rather than the entire image, reducing computational load while maintaining detection accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts and processes only the regions of interest from the full image. By identifying and isolating potential object regions first, the system avoids unnecessary computation on background areas, significantly reducing device resource consumption while preserving detection performance.

Inventive Principle:
Principle #2Taking out (Extraction)

2Measurement precision

If a large machine learning model is used for object detection, then detection accuracy is improved, but processing speed decreases

Engineering Contradiction:
Improvedetection accuracyVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The detection process is segmented into distinct phases: generating region proposals, filtering to identify regions of interest, and performing final detection. This multi-stage approach reduces the overall computational complexity and enables faster processing while maintaining accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary region proposal generation and filtering before the actual detection step. By pre-identifying and selecting only the most relevant regions, the system reduces the workload for the final detection phase, thereby improving processing speed without sacrificing detection accuracy.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If a large machine learning model is used for object detection, then detection capability is improved, but real-time performance deteriorates

Engineering Contradiction:
Improvedetection capabilityVSAvoidreal-time performance
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent extracts and processes only the essential regions of interest rather than analyzing the entire image. This extraction approach significantly reduces processing time while maintaining detection capability, enabling real-time performance on resource-constrained devices.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The system performs preliminary filtering to identify regions of interest before executing the full detection algorithm. This preliminary action reduces the amount of data requiring intensive processing, thereby reducing latency and achieving real-time detection performance.

Inventive Principle:
Principle #10Preliminary action

4Measurement precision

If a large machine learning model is used for object detection, then model performance is improved, but storage requirements increase

Engineering Contradiction:
Improvemodel performanceVSAvoidstorage requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent employs a segmented architecture where the detection process is divided into multiple specialized components (region proposal network, region of interest selector, detection head). This segmentation allows for more efficient memory utilization and reduces overall storage requirements compared to a monolithic large model.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP3493106B1Optimizations for dynamic object instance detection, segmentation, and structure mapping
Publication Date: 2023.05.03 META PLATFORMS INC
  • EP3493106B1 patent drawingFigure 1A
  • EP3493106B1 patent drawingFigure 1B
  • EP3493106B1 patent drawingFigure 2

AI summary

In one embodiment, a system may access an image and generate a feature map for the image using a neural network. The system may identify regions of interest in the feature map. Regional feature maps may be generated for the regions of interest, respectively. Each of the regional feature maps has a first, a second, and a third dimension. The system may generate a first combined regional feature map by combining the regional feature maps. The combined regional feature map has a first, a second, and a third dimension. The system may generate a second combined regional feature map by processing the first combined regional feature map using one or more convolutional layers. The system may generate, for each of the regions of interest, information associated with an object instance based on a portion of the second combined regional feature map associated with that region of interest.